When COVID-19 struck, scientists benefited from decades of prior coronavirus research. The next pandemic may not offer that head start, so a global coalition—including NVIDIA, Google DeepMind, and EMBL-EBI—has released predicted 3D structures for protein complexes from over 2,800 viruses. The open dataset, available through the AlphaFold Database, was generated using AlphaFold2 with optimization from NVIDIA's BioNeMo inference runtime, allowing large-scale prediction of viral proteomes.
About 30% of the protein interactions in the dataset are completely new to science, showing shapes never documented in the Protein Data Bank. This could provide fresh targets for vaccines and drugs, as many viral functions depend on protein complexes rather than single molecules. The predictions are labeled by confidence, and researchers can verify high-confidence hits experimentally.
The collaboration spans public health and research institutions, and the release coincides with a UN General Assembly meeting on pandemic preparedness. With an estimated 50% chance of a pandemic as severe as COVID-19 by 2050, the goal is to stockpile structural knowledge ahead of time. As one collaborator noted, the dataset gives PhD students today high-quality structural data that was unavailable to previous generations, accelerating fundamental science and hypothesis generation.